T.C.
KONYA TEKNÄ°K ÃœNÄ°VERSÄ°TESÄ°
LİSANSÜSTÜ EĞİTİM ENSTİTÜSÜ
KONYA TEKNÄ°K ÃœNÄ°VERSÄ°TESÄ°
LİSANSÜSTÜ EĞİTİM ENSTİTÜSÜ
Automated Detection and Analysis for Diagnosis in Cephalometric X-ray Image Using Convolutional Neural Network¶
This project is build on top of the Wang Cwei dataset which can be found in the
link: https://figshare.com/s/37ec464af8e81ae6ebbf
The scope of this work is to read the images and display the dots annotated by a professional medical doctor¶
In [46]:
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import os
from skimage import io, transform, img_as_float
from sklearn.model_selection import train_test_split
import torch
import torch.nn as nn
import torchvision.datasets as dsets
from torchvision.transforms import ToTensor
from torchvision import datasets, transforms, models # add models to the list
from torch.utils.data import Dataset, DataLoader, TensorDataset
from torchvision import models
from torch.autograd import Variable
import torch.nn.functional as F
from torchvision.utils import make_grid
import time
import random
# ignore harmless warnings
import warnings
warnings.filterwarnings("ignore")
Lets perfom this operation on one image and sclale before performing a similar operation on multiple images
In [91]:
# Take a look at one of the image samples and labels
#NOTE: THE IMAGE FOLDERS HAS BEEN MODIFIED AND SEPERATED INTO TRAIN AND TEST FOLDERS SETS
SAMPLE_PATH = "data/RawImage/Train/TrainingData/005.bmp"
TXT_PATH = "data/AnnotationsByMD/400_senior/005.txt"
# import sample image
img = io.imread(SAMPLE_PATH, as_gray=True)
img
Out[91]:
In [92]:
img.shape
Out[92]:
In [93]:
# import sample coordinates from text as tuples
def extract_labels_from_txt(path):
with open(path, "r") as f:
# only first 19 are actual coords in dataset label files
coords_raw = f.readlines()[:19]
coords_raw = [tuple([int(float(s)) for s in t.split(",")]) for t in coords_raw]
return coords_raw
In [94]:
coords_raw = extract_labels_from_txt(TXT_PATH)
coords_raw
Out[94]:
In [95]:
plt.rcParams["figure.figsize"] = [32,18]
plt.style.use(['dark_background'])
fig = plt.figure()
ax1 = fig.add_subplot(2, 2, 1)
ax2 = fig.add_subplot(2, 1, 1)
ax1.imshow(img, cmap="gray")
# also plot resized image for later
orig_y, orig_x = img.shape[:2]
SCALE = 15
# for rescale, use same target for both x&y axis
rescaled_img = transform.resize(img,(orig_y/SCALE,orig_y/SCALE))
ax2.imshow(rescaled_img, cmap="gray")
for c in coords_raw:
# add patches to original image
# could also just plt.scatter() but less control then
ax1.add_patch(plt.Circle(c, 5, color='r'))
# and rescaled marks to resized images
x,y = c
x = int(x*(orig_y*1.0/orig_x)/SCALE)
y = int(y/SCALE)
ax2.add_patch(plt.Circle((x,y), 1, color='g'))
plt.show()
In [96]:
def print_image(img,labels):
print(img.shape)
plt.rcParams["figure.figsize"] = [32,18]
fig = plt.figure()
ax1 = fig.add_subplot(2, 2, 1)
ax2 = fig.add_subplot(2, 1, 1)
ax1.imshow(img, cmap="gray")
# also plot resized image for later
orig_y, orig_x = img.shape[:2]
SCALE = 15
# for rescale, use same target for both x&y axis
rescaled_img = transform.resize(img,(orig_y/SCALE,orig_y/SCALE))
ax2.imshow(rescaled_img, cmap="gray")
for c in coords_raw:
# add patches to original image
# could also just plt.scatter() but less control then
ax1.add_patch(plt.Circle(c, 5, color='r'))
# and rescaled marks to resized images
x,y = c
x = int(x*(orig_y*1.0/orig_x)/SCALE)
y = int(y/SCALE)
ax2.add_patch(plt.Circle((x,y), 1, color='g'))
plt.show()
In [98]:
BASE_IMAGE_PATH='data/RawImage/Train/TrainingData/'
BASE_CORD_PATH='data/AnnotationsByMD/400_senior/'
def display_image_and_cord(image_number,img_path, cord_path):
data = []
target = []
for i, fi in enumerate(os.listdir(img_path)):
if i<image_number:
loop_img = io.imread(img_path + fi, as_gray=True)
lf = fi[:-4] + ".txt"
loop_labels = extract_labels_from_txt(cord_path + lf)
loop_labels = (np.array(loop_labels))
print(loop_img)
print_image(loop_img,loop_labels)
display_image_and_cord(10,BASE_IMAGE_PATH, BASE_CORD_PATH)
Define transforms¶
In the previous section we looked at a variety of transforms available for data augmentation (rotate, flip, etc.) and normalization.
Here we'll combine the ones we want, including the recommended normalization parameters for mean and std per channel.
In [99]:
train_transform = transforms.Compose([
transforms.Resize(224), # resize shortest side to 224 pixels
transforms.CenterCrop(224), # crop longest side to 224 pixels at center
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
])
test_transform = transforms.Compose([
transforms.Resize(224),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
])
In [100]:
root = 'data/RawImage/'
train_data = datasets.ImageFolder(os.path.join(root, 'Train'), transform=train_transform)
test_data = datasets.ImageFolder(os.path.join(root, 'Test'), transform=test_transform)
In [101]:
train_data
Out[101]:
In [102]:
test_data
Out[102]:
In [103]:
torch.manual_seed(42)
train_loader = DataLoader(train_data, batch_size=10, shuffle=True)
test_loader = DataLoader(test_data, batch_size=10, shuffle=True)
In [104]:
class_names = test_data.classes
print(class_names)
print(f'Training images available: {len(train_data)}')
print(f'Testing images available: {len(test_data)}')
Display a batch of images¶
To verify that the training loader selects cat and dog images at random, let's show a batch of loaded images.
Recall that imshow clips pixel values <0, so the resulting display lacks contrast. We'll apply a quick inverse transform to the input tensor so that images show their "true" colors.
In [105]:
# Grab the first batch of 10 images
for images,labels in train_loader:
break
In [106]:
images
Out[106]:
In [107]:
labels
Out[107]:
In [108]:
# Print the labels
im = make_grid(images, nrow=4) # the default nrow is 8
# Inverse normalize the images
inv_normalize = transforms.Normalize(
mean=[-0.485/0.229, -0.456/0.224, -0.406/0.225],
std=[1/0.229, 1/0.224, 1/0.225]
)
im_inv = inv_normalize(im)
# Print the images
plt.figure(figsize=(32,18))
plt.imshow(np.transpose(im_inv.numpy(), (1, 2, 0)));
